Select Page

Weak AI, Narrow AI

AI which has been trained to work well for a particular type of problem – for example text to speech. Contrast with Strong AI

Strong AI, Artificial General Intelligence

A theoretical form of AI which is able to generalize to and work well for any type of problem, like humans do

Computational Graph

A flowchart like structure which defines how inputs are processed in steps to provide a result. TensorFlow, PyTorch and CoreML are examples of libraries which operate on computational graphs

Training, Optimization

Training and Optimization refer to the iterative process through which a model is refined.

Gradient Decent

Gradient Decent is a training method which searches for a set of common factors which allow the model to provide the correct answer for any input supplied, by formulating a map where the best factors are in wells and worse answers are on hills. The best factors are found by moving through this map to find the lowest points

Knowledge Graph

A collection of entities and concepts and their relations stored in a connected graph structure like a brainstorm

Forward Chaining

Formulating a logical conclusion when given a set of facts (e.g. Fred is green and croaks – so Fred must be a Frog)

Backward Chaining

Deriving a set of logical facts given a statement (e.g. Fred is a Frog – so Fred must also croak)

Annotation, Label, Tag

An annotation/label/tag is a minimal description of some concept, entity or location contained within data used with a model

Bounding Box

A sub-region within some data like part of an image or an audio clip which relates to a specific annotation

Dependency tree

An output produced by natural language processing which highlights the relationships between individual words in some analyzed text


Introduced during the training process through prejudices which may exist in the source material, which the in turn results in the model displaying more of the same

Deep Learning

Using multiple layers within a computation graph to iteratively learn and exploit higher level abstractions during decision making


Commonly used in reference to chatbots, they map spoken commands to programatic responses depending on the users intended goal

Machine Translation

AI which converts one token stream into another, for example from English to French


When a model has been trained in a manner which does not generalize, so it can only operate accurately when presented with data similar to what it was presented during training

Supervised Learning

Training a model using source material which was previously annotated

Unsupervised Learning

Training a model without explicit prior annotation

Reinforcement Learning

Training a model by way of positive or negative feedback in response to an action taken

Transfer Learning

Repurposing or specializing a previously trained model to operate accurately on new, previously unseen data, making use of learnings from previous training (e.g. Presenting new images to a visual recognition model, the model can reuse previously learned edges and small shapes)

Search, understand
and simplify your workflows.

Ready to see how Cyte can help you?

Operating System: Mac OSX 10.15+ or Windows 10+ | Minimum Memory: 8 GB RAM

Minimum Free HDD Space: 8 GB | Minimum CPU Spec: 1.9 GHz Quad Core | Version: 0.7.2 (Release notes)

About         Features         Help Centre         App Feedback         Contact         Blog